I built an AI calorie tracker that admits when it's guessing

I'm bk, a software developer in Toronto. I just shipped my first solo iOS app, CalScore, an AI calorie and macro tracker. This post is about the one product decision I care most about, and I'd love feedback from other devs.

The problem: AI calorie numbers look more precise than they are

Most photo-based calorie apps take a picture of your plate and return one number, like "612 kcal". It looks exact, but a photo can't see the oil in the pan, how deep the bowl is, or what's hiding under the sauce. When the estimate is wrong, nothing tells you, and once you notice, you stop trusting the app.

The decision: show the uncertainty

So in CalScore, every photo scan returns:

• estimated calories, protein, carbs and fat
• a calorie range instead of a single number
• a confidence level for the estimate

When confidence is low, the app nudges you to double-check the portion, scan a barcode, or search the food database instead. Barcode and search use USDA FoodData Central and Open Food Facts, for when you want precision rather than an estimate.

One number for the whole day: Fuel Score

Instead of six charts, there's a daily Fuel Score (0–100) that combines calories, protein, hydration and macro balance, so you can tell at a glance whether you're on track.

Free vs Pro (being upfront)

• Free: calorie and macro targets, water tracking and reminders, browsing the food database, and the daily Fuel Score.
• Pro (weekly, monthly or annual, with a free trial on annual): AI photo scan, barcode scanner, food logging, custom foods, an AI nutrition coach, weight log and trends, no ads.

I'm actively considering making barcode and manual logging free, which is one of the things I'd like your opinion on.

What I'd love feedback on

1. Is a calorie range plus a confidence level more useful than a single number, or just noise?
2. Is the free/Pro split fair? What would you expect to be free in a tracker like this?
3. For those who've built on vision models: how do you communicate uncertainty to users without making the product feel unreliable?

It's iPhone-only (iOS 17+) and brand new, with zero ratings so far, so honest feedback is very welcome:
https://apps.apple.com/ca/app/calscore-ai-calorie-tracker/id6827339697

Estimates are for general wellness, not medical advice.

Disclosure: this post was written and published by an AI assistant on my behalf. I'm the developer of CalScore.